Ontology-Based Recommendation of Editorial Products

TL;DR

This paper presents an ontology-based recommender system that leverages a large-scale research taxonomy to assist publishers in selecting relevant scientific products for specific venues, improving over manual methods.

Contribution

The paper introduces the Smart Book Recommender, an innovative ontology-based system utilizing the Computer Science Ontology for precise, automated editorial product recommendations.

Findings

Effective recommendation accuracy confirmed by user evaluation.

Supports exploration of why items are recommended via interactive graph.

Enhances editorial decision-making with semantic analysis.

Abstract

Major academic publishers need to be able to analyse their vast catalogue of products and select the best items to be marketed in scientific venues. This is a complex exercise that requires characterising with a high precision the topics of thousands of books and matching them with the interests of the relevant communities. In Springer Nature, this task has been traditionally handled manually by publishing editors. However, the rapid growth in the number of scientific publications and the dynamic nature of the Computer Science landscape has made this solution increasingly inefficient. We have addressed this issue by creating Smart Book Recommender (SBR), an ontology-based recommender system developed by The Open University (OU) in collaboration with Springer Nature, which supports their Computer Science editorial team in selecting the products to market at specific venues. SBR recommends books, journals, and conference proceedings relevant to a conference by taking advantage of a semantically enhanced representation of about 27K editorial products. This is based on the Computer Science Ontology, a very large-scale, automatically generated taxonomy of research areas. SBR also allows users to investigate why a certain publication was suggested by the system. It does so by means of an interactive graph view that displays the topic taxonomy of the recommended editorial product and compares it with the topic-centric characterization of the input conference. An evaluation carried out with seven Springer Nature editors and seven OU researchers has confirmed the effectiveness of the solution.